Skip to content

Playbooks · Team Standards

AI polish hides the flaws reviewers should catch. Label your drafts early and write the team's quality bar down

Atlassian found reviewers were 22% less likely to catch a critical flaw in AI-polished drafts, and a two-word label almost cancelled the effect.

by ·5 min read·

New here? Start with the free AI Survival Kit →

AI polish hides the flaws reviewers should catch. Label your drafts early and write the team's quality bar down

0:00 / 7:09

This voice is generated by AI.

AI polish hides the flaws reviewers should catch. Label your drafts early and write the team's quality bar down

Atlassian's Teamwork Lab handed 903 knowledge workers the same draft proposal with two planted flaws: no plan to get employees to the new self-help hub, and a success metric that counted website traffic instead of fewer IT tickets. Reviewers who received the AI-polished version were 22% less likely to flag the first flaw and 15% less likely to catch the second. They spent 62% more time reviewing it and were 18% less willing to give feedback. Clean prose told them the work was finished. ZipRecruiter's 2026 AI Worker Report, a September survey of more than 1,500 US workers, found 40.9% use AI to draft and edit text, and 49.9% report higher productivity expectations. More workers are polishing drafts with AI, and in Atlassian's test polished drafts got less scrutiny.

Polish reads as finished

The polished draft earned its author credit. Reviewers were 1.7 times more likely to say the author put in high effort and 2.5 times more likely to rate the draft as highly finalised. That is the trap. The signal that wins approval is the same signal that switches off scrutiny.

Both flaws in the test were structural: a missing adoption plan and the wrong measure of success. A model fixes grammar in seconds. It will not tell you that you are counting the wrong thing.

Who missed the flaws matters. Gen Z and entry-level reviewers saw steep drops in flaw detection. For Gen X reviewers and managers, polish made no meaningful difference. Atlassian calls experience a firewall. The reviewer who finds the missing adoption plan inside a well-formatted document is doing the part of the job a model does not do.

The fix in the study was two words. An AI-polished draft marked "Status: Early Draft", paired with an explicit request for feedback, almost entirely cancelled the effect. Perceived finality fell from 37% to 21%. Willingness to critique returned to baseline. High-effort ratings went up, from 64% to 69%. The author kept the credit and got the critique.

Workslop is a missing standard

On 27 October Atlassian published a second study. It defines workslop as AI-generated work that looks polished but lacks the context, judgement and substance needed to move work forward, which shifts the interpreting, checking and rewriting onto whoever receives it. The Teamwork Lab's position is that this is a standards problem. Its researchers write that careless and conscientious people alike produce workslop when they lack guidance.

To test that, 31 employees from five teams in Atlassian's Workplace group sat a 60-minute workshop in June 2026. It ran in four phases: what slop looks like, what a good page looks like, the shared principles behind a good page, and the strategies that keep quality high.

Before the session, 62% agreed they knew what their team considered a high-quality page. Afterwards, 100% did. Agreement that they knew how to use AI to produce one rose from 60% to 97%. 94% said they felt more comfortable giving feedback on AI-generated pages. Two weeks later, agreement that the team's pages met a high standard had risen from 40% to 67%. Atlassian states plainly that this last result is not statistically significant, given the size of the pilot. Treat it as a pattern worth testing on your own team.

Skipping the standard has a price. Atlassian cites a 2025 BetterUp and Stanford survey in which 42% of people who received workslop saw the sender as less trustworthy and 37% saw them as less intelligent. Polish buys effort credit on the day and spends trust afterwards.

One tactic from the workshop is worth lifting whole. Participants planned to record a short Loom talking through their thinking before asking Rovo, Atlassian's AI assistant, to draft the page. Reasoning first, model second. AI does the routine work. You do the thinking. The order is the point.

Your Next Move

  1. Label every AI-polished draft you send for review this month. Put the status and the one question you need answered on the first line: "Status: early draft. Is the success metric right?" Atlassian's own advice is that when you want hard critique, sending the unpolished version is faster for everyone.

  2. Run the 60-minute standards session before your next planning cycle. Follow Atlassian's four phases: what slop looks like, what a good page looks like, the shared principles, and the strategies that keep quality high. Bring one real example of slop and one page the team agrees is good. Write the principles down where everyone, and every prompt, can reference them.

  3. Read for structure before you read for prose. When a polished document lands, ask two questions first: how does this reach the people it is for, and does the metric measure the outcome? If you are senior, review one document alongside a junior colleague this month and say out loud which flaws you are hunting. The study shows experience holds up against polish. Pass it on deliberately.

I build systems like the one publishing this site. → Work with me

About the author

Jo

Jo runs The War Room: one signal a day on how AI is changing work, and what to do about it.

Sources

Get the Briefing

Want more intelligence like this?

The weekly briefing, free — and the AI Survival Kit with it.

Takes 30 seconds. No spam.

More in Playbooks

See all